An emergency sourcing method, equipment, and medium for the biopharmaceutical supply chain.

Abstract

This specification discloses an emergency sourcing method, device, and medium for the biopharmaceutical supply chain, relating to the field of supply chain technology. The method includes: receiving a quality deviation emergency request containing a deviation batch identifier and deviation type; based on the quality deviation emergency request, determining the deviation batch and the corresponding set of affected batches in a pre-constructed dynamic supply chain knowledge graph; acquiring corresponding production-related data and logistics-related data; inputting the production-related data and logistics-related data into a pre-trained causal risk assessment model; determining the risk transmission path information from the deviation batch to the affected batches in the set of affected batches; calculating the risk index for each affected batch to determine the batch disposal decision tree for the corresponding disposal instruction; and based on the batch disposal decision tree and a pre-stored supplier profile database, determining the corresponding supplier selection scheme and material allocation scheme to generate an emergency purchase order, thereby achieving emergency sourcing.

Description

Technical Field

[0001] This specification relates to the field of supply chain technology, and in particular to an emergency sourcing method, equipment and medium for the biopharmaceutical supply chain. Background Technology

[0002] In the biopharmaceutical industry, the security and reliability of the supply chain are paramount, affecting not only the normal operation of companies but also directly impacting patient safety and drug efficacy. The biopharmaceutical supply chain is characterized by its complex structure, stringent regulations, high sensitivity to material quality, and hidden risk transmission pathways. Therefore, rapid response, accurate traceability, and emergency source identification of quality deviations occurring during production or logistics are core challenges in ensuring supply chain resilience.

[0003] Currently, various technological solutions exist in the field of supply chain risk management. One existing solution constructs a network model of the supply chain and uses graph algorithms to analyze the relationships between nodes, thereby identifying potential single risk points or risk propagation paths. Another existing solution integrates multi-source data to construct a knowledge graph and uses graph neural networks for dynamic risk prediction and anomaly detection. These technologies have made progress in improving the breadth of supply chain visualization and risk identification. However, when these existing technologies are applied to the specific scenario of biopharmaceuticals, which are subject to high regulation and high standards, they often lack the ability to deeply and structurally integrate biopharmaceutical-specific data (such as continuous environmental monitoring time-series data and strict compliance constraints) and business relationships (such as batch-level material traceability and cleanroom equipment sharing). This results in the graphs or network models that are difficult to accurately depict the key causal chains of how quality deviations are transmitted between batches through specific production equipment, operating procedures, or storage environments. Secondly, when it is necessary to find emergency alternative sources for disrupted supply, the decision support functions of existing technologies often remain at the level of providing risk warnings or screening supplier lists, failing to meet the requirement of generating directly executable emergency sourcing solutions within the emergency time window.

[0004] Therefore, in responding to quality deviations in the biopharmaceutical supply chain and conducting emergency source tracing, it cannot meet the biopharmaceutical industry's requirements for rapid risk response, accurate decision-making, and reliable compliance. Summary of the Invention

[0005] This specification provides one or more embodiments of an emergency sourcing method, device, and medium for the biopharmaceutical supply chain, which addresses the following technical problem: in the process of responding to quality deviation events in the biopharmaceutical supply chain and conducting emergency sourcing, it is impossible to meet the biopharmaceutical industry's requirements for risk response speed, decision-making accuracy, and compliance reliability.

[0006] One or more embodiments of this specification employ the following technical solutions: This specification provides one or more embodiments of an emergency sourcing method for a biopharmaceutical supply chain. The method includes: receiving a quality deviation emergency request containing a deviation batch identifier and a deviation type; based on the quality deviation emergency request, determining the deviation batch and the corresponding set of affected batches in a pre-constructed dynamic supply chain knowledge graph; acquiring corresponding production-related data and logistics-related data, wherein the production-related data includes time-series data from production environment sensors and production operation records, and the logistics-related data includes time-series data from transportation vehicle sensors and warehousing environment records; inputting the production-related data and the logistics-related data into a pre-trained causal risk assessment model to determine the risk transmission path information from the deviation batch to the affected batches in the set of affected batches; calculating the risk index for each affected batch to determine a batch disposal decision tree for the corresponding disposal instruction; and based on the batch disposal decision tree and a pre-stored supplier profile database, determining the corresponding supplier selection scheme and material allocation scheme to generate an emergency purchase order and achieve emergency sourcing.

[0007] This specification provides one or more embodiments of an emergency sourcing device for the biopharmaceutical supply chain, comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the above-described method.

[0008] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0009] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: By using a dynamic supply chain knowledge graph, data on materials, equipment, and the environment scattered across different systems are linked together. In the event of a deviation, all batches linked through production and logistics can be automatically and comprehensively traced, overcoming the ambiguity and lag of traditional manual investigation. A pre-trained causal risk assessment model is introduced to quantify risk transmission analysis. The model calculates risk transmission path information and risk indicators, transforming the binary judgment of whether an item is affected into a specific, quantifiable risk value for each batch. This allows subsequent handling to be based on objective data rather than subjective experience, achieving resource allocation based on precise risk measurement. Furthermore, a sourcing plan is generated based on a quantified batch handling decision tree and a supplier profile database, achieving intelligent connection from risk analysis to business action. This ensures that the starting point for sourcing decisions is the precise quantification of risk gaps, and inherently integrates the supplier's historical performance and compliance status (from the profile database). This allows emergency procurement not only to fill material gaps but also to proactively leverage the results of previous risk analysis, selecting supply sources that truly reduce system risk. Through these technical solutions, precise containment and scientific recovery can be implemented as quickly as possible after a quality deviation occurs, maximizing compliance with regulatory requirements and minimizing business interruption losses. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating an emergency sourcing method for the biopharmaceutical supply chain provided in the embodiments of this specification; Figure 2 This is a schematic diagram of an emergency sourcing device for the biopharmaceutical supply chain, provided as an embodiment of this specification. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0012] This specification provides an emergency sourcing method for the biopharmaceutical supply chain. It should be noted that the executing entity in this specification can be a server or any device with data processing capabilities. Figure 1 A flowchart illustrating an emergency sourcing method for the biopharmaceutical supply chain provided in this specification is shown below. Figure 1 As shown, the main steps include the following: Step S101: Receive a quality deviation emergency request containing a deviation batch identifier and deviation type. Based on the quality deviation emergency request, determine the deviation batch and the corresponding set of affected batches in the pre-built dynamic supply chain knowledge graph, and obtain the corresponding production-related data and logistics-related data.

[0013] The production-related data includes time-series data from production environment sensors and production operation records, while the logistics-related data includes time-series data from transportation vehicle sensors and warehousing environment records. In one embodiment of this specification, before determining the batch with quality deviation and the corresponding set of affected batches in a pre-built dynamic supply chain knowledge graph based on the quality deviation emergency request, the method further includes constructing the dynamic supply chain knowledge graph.

[0014] First, data on material batch entities, production process entities, equipment entities, and their supply, processing, and location relationships are extracted from the Enterprise Resource Planning (ERP) database, Manufacturing Execution System (MES) database, and Warehouse Management System (WMS) database as the primary data source for the graph. Specifically, structured data is extracted periodically or triggered from the ERP database, MES database, and WMS database through pre-configured data connectors and API interfaces. The Enterprise Resource Planning (ERP) system database stores purchase orders, material master data, and supplier information. Material batch entities extracted from this database refer to a specific set of raw materials, intermediates, or finished products with unique identifiers (such as batch numbers). Their attributes include material code, production date, expiration date, and supplier code. The Manufacturing Execution System (MES) database records detailed operational information from the production site. Production process entities extracted from this database represent specific processing steps (such as weighing, mixing, and filling). Their attributes include process code, standard operating procedure number, and execution station. Equipment entities extracted from this database refer to machines or devices used for production (such as reactors, filling machines, and freeze dryers). Their attributes include equipment number, model, and status. Supply relationship data extracted simultaneously from these systems characterizes the input-output sequence between material batches (e.g., material batch A is processed through process X to produce material batch B). Processing relationship data characterizes which equipment entity executed which production process entity for which material batch entity at which time. Location relationship data characterizes which storage location (e.g., cold storage area, inspection area) a material batch entity is located in at a specific time.

[0015] From the IoT platform database, temperature sensor entities, differential pressure sensor entities, and particle counter entities associated with the device entity and storage space entity are accessed, and their generated environmental monitoring time-series data are acquired in real time as a second graph data source. Specifically, secondly, sensor data streams are accessed in real time from an independent IoT platform database via a message queue (such as Kafka). This platform manages physical sensors deployed on production equipment (such as the chambers of incubators) and storage spaces (such as cold storage shelves). Temperature sensor entities, differential pressure sensor entities, and particle counter entities are digital mappings of these physical devices, each entity having a unique identifier and being associated with a specific device entity or storage space entity through installation relationships. The system receives and caches the environmental monitoring time-series data sent by these sensors in real time, i.e., a sequence of timestamped measurements (such as degrees Celsius, Pascals, and particles per cubic meter) sampled at a fixed frequency. This real-time, high-frequency time-series data constitutes the second graph data source, providing continuous signals reflecting the state of the physical world for the knowledge graph.

[0016] Then, using an entity linking algorithm, the material batch entities and equipment entities in the first knowledge graph data source are aligned and merged with the corresponding sensor entities in the second knowledge graph data source to generate entity relationship pairs. Specifically, firstly, entity recognition is performed, extracting features from the equipment entity description information (such as equipment number, name, and location) in the first knowledge graph data source and the sensor metadata (such as installation location description and equipment number) in the second knowledge graph data source; then, similarity calculation is performed, using a combination of rule-based (such as exact number matching) and text similarity model to calculate the matching score between equipment entities and sensor entities; finally, alignment and merging are performed, and for entity pairs with matching scores exceeding a threshold, a unified equipment node is created in the knowledge graph. This node simultaneously carries attributes from the business system (such as equipment model) and associations from the IoT system (such as the list of owned sensors), and the relationship between this unified node and related material batches and production processes is established. For example, merge the "filling machine-B-07" entity in the ERP with the "vibration sensor-V-007" entity in the Internet of Things, and establish the processing relationship between this unified equipment node and the "product batch number 20231025A" batch entity filled by it at a specific time point.

[0017] Then, this entity relationship pair is associated with the Good Manufacturing Practice (GMP) clause entities obtained from the regulatory database, attaching compliance constraint attributes to the production process relationship edges in the graph. Structured GMP clause entities are retrieved from the regulatory database; these entities represent the specific requirements of the regulations in a machine-readable form (e.g., clause entity - temperature requirement: storage temperature should be between 2-8°C). Using predefined mapping rules, these clause entities are associated with the relevant production process relationship edges in the graph. For example, a compliance constraint attribute is attached to the processing relationship edge between the filling process and batch number 20231025A. This attribute might point to the clause entity - aseptic operation requirement and contain specific parameters of the constraint (e.g., environmental particle count should be lower than the dynamic Class A standard). This process essentially transforms textual regulatory requirements into associative and reasonable graph attributes.

[0018] Finally, the entities and relationships with the compliance constraint attributes are stored in a graph database to form this dynamic supply chain knowledge graph. All entities (such as material batches, equipment, sensors, and regulatory clauses) and relationships (such as supply, processing, monitoring, and being bound) with compliance constraint attributes generated in the above steps are persistently stored in a graph database (such as Neo4j) as nodes and edges. Each node and edge has a unique identifier, type label, attribute key-value pairs, and timestamp information, thus forming a dynamic supply chain knowledge graph that can be dynamically updated and incorporates business logic, real-time status, and regulatory constraints.

[0019] Conventional methods often involve simple correlation queries on business system data or the creation of static supplier-material correspondence tables, resulting in severe data silos, weak real-time perception capabilities, and compliance checks typically treated as a separate post-audit step. The embodiments in this specification, by constructing a deeply integrated dynamic knowledge graph, firstly achieve real-time and accurate mapping between business flows and physical flows, directly binding abstract material batches to specific production equipment and real-time environmental sensor data. This allows any quality deviation to be quickly located to specific physical equipment, spatial location, and precise time period, significantly improving the granularity and speed of risk tracing. Secondly, by transforming regulatory clauses into associative knowledge graph entities and dynamically linking them to the production process, compliance requirements are structurally embedded in production operations and monitored in real time. This transforms compliance from a passive audit item into a proactive, traceable knowledge graph attribute, significantly reducing systemic risks caused by compliance oversights. Finally, this deeply integrated graph structure provides a unique and complete data foundation for subsequent risk transmission analysis based on graph algorithms and emergency sourcing based on multi-objective optimization (considering real-time risks, recertification costs and time). This enables a full understanding of the complex multi-level dependencies, real-time status and hard regulatory constraints in the biopharmaceutical supply chain, thereby generating practical and optimal emergency solutions.

[0020] In one embodiment of this specification, upon receiving a quality deviation emergency request containing a deviation batch identifier and a deviation type, the quality deviation batch and the corresponding set of affected batches are determined based on the quality deviation emergency request in a pre-built supply chain knowledge graph. This is specifically achieved in the following manner: First, based on the deviation batch identifier, locate the corresponding deviation batch entity node in the dynamic supply chain knowledge graph, and parse the deviation batch identifier in the request. This identifier is usually a unique batch number or code bound to the production record. Using this identifier as the key, perform a query in the graph database (such as Neo4j) to locate the corresponding deviation batch entity node in the knowledge graph. This node carries all the graph-based attributes of the batch.

[0021] Next, based on the deviation type, a corresponding graph traversal strategy is matched. This graph traversal strategy defines the time range threshold and the set of relationship types for querying entity relationships within the dynamic supply chain knowledge graph. According to the deviation type in the request (such as microbial contamination, visible foreign matter, or temperature exceeding limits), the corresponding strategy is matched from a pre-configured graph traversal strategy library. The graph traversal strategy library is predefined, mapping different deviation types to different graph exploration rules. Each graph traversal strategy explicitly defines two types of core parameters: one is the time range threshold. For example, for microbial contamination that may spread through the air, the strategy might specify traversing back to all related batches within a specific time period before and after the start time of the batch with the deviation; for temperature deviations, it might focus on all batches on the same cold chain transport vehicle. Second, the set of relationship types indicates which types of relationship edges should be focused on when traversing the graph. For example, for the risk of cross-contamination, the strategy will include shared equipment processing relationship edges (i.e., batches using the same key equipment such as filling machines) and spatiotemporal proximity relationship edges (i.e., batches located in the same cleanroom area within a similar time period); for raw material issues, the strategy will focus on supply relationship edges (i.e., batches with direct raw material input-output relationships).

[0022] Then, according to the graph traversal strategy, starting from the deviation batch entity node, the graph is traversed along supply relationship edges, shared equipment processing relationship edges, and spatiotemporal proximity relationship edges to filter out all batch entity nodes within the time range threshold. Based on the matched graph traversal strategy, starting from the located deviation batch entity node, the graph is traversed in the graph database along the set of relationship types specified by the strategy. The traversal algorithm (such as breadth-first search) traces upstream raw material batches and downstream product batches along supply relationship edges; finds all batch nodes that have used the same key equipment for processing along shared equipment processing relationship edges; and searches for batch nodes processed or stored in the same or adjacent physical locations (such as the same cleanroom) within the time window specified by the strategy along spatiotemporal proximity relationship edges. This process filters out all batch entity nodes within the time range threshold, forming a set of potential impacts based on production and spatial relationships.

[0023] Next, the risk transmission in the logistics process is further considered. All selected batch entity nodes are joined with in-transit batch entity nodes connected by logistics linkage edges to form an initial set of affected batches. In the graph, all in-transit or in-stock batch entity nodes connected to the current deviation batch node by logistics linkage edges (such as the same transport vehicle or the same warehouse shelf) are identified. These nodes are joined with the production-related batch node set obtained in the previous step to form the initial set of affected batches, ensuring that batches that have not yet entered the production process but are already in the logistics chain and may be subject to physical contamination or environmental impact are also considered.

[0024] Finally, to improve the accuracy of risk assessment and avoid unnecessary handling of qualified batches that have already passed inspection, filtering based on real-time quality data is implemented. The key product quality attribute values ​​of each batch entity node in the initial affected batch set are extracted from the dynamic supply chain knowledge graph and compared with a preset quality standard range. Batch entity nodes whose key quality attribute values ​​fall within the quality standard range are removed, generating the final affected batch set. Specifically, the key product quality attribute values ​​associated with each batch entity node in the initial affected batch set are extracted from the dynamic supply chain knowledge graph. These attribute values ​​(such as sterility test results, endotoxin content, and visible foreign matter count) originate from the database of the integrated laboratory information management system and are attached to the corresponding batch nodes as attributes. These measured values ​​are automatically compared with a preset quality standard range (such as pharmacopoeia standards or internal company standards). Only when all relevant key quality attribute values ​​of a batch are within the quality standard range is the batch considered compliant and removed from the set. After this round of screening, the remaining batch nodes constitute the final set of affected batches. The batches in this set all have a clear risk exposure association and their quality status has not been verified as qualified, providing a precise target range for subsequent refined risk assessment and handling.

[0025] Conventional methods heavily rely on human experience, with quality personnel manually checking potentially affected batches based on memory or limited paper records. This process is tedious, time-consuming, and prone to incomplete checks or overestimation of the impact due to information omissions or highly complex correlations. The embodiments in this specification, through automated strategy traversal based on dynamic knowledge graphs, achieve a systematic and comprehensive exploration of multi-dimensional risk transmission paths. It not only traces material flows along the forward and reverse directions of the supply chain but also intelligently identifies risks arising from hidden paths such as shared equipment, shared spaces, and shared logistics resources. This multi-dimensional correlation is difficult for humans to fully grasp. Meanwhile, the introduction of real-time key quality attribute data for automated filtering upgrades the traditional extensive model of first locking all batches and then testing them one by one to an intelligent model that first locks based on correlations and then focuses precisely on real-time data. This greatly improves the accuracy and efficiency of determining the scope of impact, saving a significant amount of manpower and testing costs. Furthermore, by quickly and accurately identifying the truly risky batches, it buys valuable time for taking timely and correct containment and disposal measures, and avoids resource waste or risk spread caused by misjudgment of the scope. This significantly enhances the resilience and scientific nature of the biopharmaceutical supply chain in responding to quality emergencies.

[0026] Step S102: Input production-related data and logistics-related data into the pre-trained causal risk assessment model to determine the risk transmission path information from the deviation batch to the affected batches in the affected batch set, calculate the risk index of each affected batch, and determine the batch disposal decision tree of the corresponding disposal instruction.

[0027] In one embodiment of this specification, the production-related data and the logistics-related data are first input into a pre-trained causal risk assessment model to determine the risk transmission path information from the deviated batch to the affected batches in the affected batch set. This is specifically achieved in the following manner: Historical deviation case data, including root cause annotations, is extracted from the historical deviation event database. This data includes historical production-related data, historical logistics-related data, and annotated root cause entities. Training data is also extracted from the database, a continuously accumulating structured knowledge base. Data sources include internal deviation investigation reports, corrective and preventative action records, and publicly available events from regulatory announcements and industry literature, structured using natural language processing. Each historical deviation case is a complete record. Historical production-related data includes environmental sensor readings (such as temperature and pressure fluctuation curves), equipment numbers, and operator information for the processes involved in the deviation batch. Historical logistics-related data includes transportation temperature records and storage location history for the batch before the deviation was discovered. The annotated root cause entities are confirmed through investigation and annotated by professionals to the root cause nodes in the knowledge graph, such as HEPA filter leaks in air conditioning systems, worn filling needles, or non-uniform crystal forms of raw materials from supplier A. These annotations establish causal relationships between specific anomalies and physical or logical entities in the knowledge graph.

[0028] Using historical deviation case data, a Bayesian network model is trained to determine the causal risk assessment model. In this model, network nodes represent different risk factor entities and deviation phenomenon entities, directed edges between nodes represent causal relationships, and edge weights represent conditional probabilities. The specific modeling process is as follows: Based on entity types related to production, logistics, and quality in the knowledge graph (such as "equipment type," "sensor type," and "deviation phenomenon"), domain experts initially define the network topology, i.e., determine the network nodes and their potential directed edge directions. For example, nodes might include "excessive vibration of filling machine," "excessive particle count in aseptic filling area," and "visible foreign matter in product." Then, parameter learning is performed using historical deviation case data. Through statistical calculations or expectation-maximization algorithms, the specific conditional probability on each edge is determined, i.e., the probability of a subsequent node (such as "visible foreign matter in product") occurring given that the preceding node (such as excessive vibration of filling machine) has occurred. The trained model encapsulates the probabilistic causal knowledge learned from historical data about how various risk factors lead to quality deviations.

[0029] When faced with new real-time deviations, reasoning is performed. The acquired production-related data and logistics-related data are transformed into input feature vectors defined by the causal risk assessment model. These vectors are then input into the pre-trained causal risk assessment model to perform probabilistic reasoning. When the entity nodes of the deviation batch are used as evidence, the posterior probability of quality deviations occurring in the entity nodes of the affected batches in the affected batch set is obtained, thus determining the path transmission strength. The acquired production-related data and logistics-related data undergo feature engineering to transform them into input feature vectors that the model can recognize. For example, time-series temperature sensor data over a period of time is converted into statistical features (mean, variance, duration exceeding threshold), and production operation records are converted into classification codes. This feature vector is input into a trained causal risk assessment model, and probabilistic inference is performed in a Bayesian network using anomalies observed by the entity nodes of the deviated batch (such as positive microbial tests) as evidence. Inference algorithms (such as connection tree algorithms) calculate the posterior probability of all other nodes in the network. Among them, the posterior probability of the quality deviation of the entity nodes of the affected batch in the set of affected batches serves as a measure of the path transmission strength of risk from the source batch to this batch. This strength value is a quantified probability value, representing the possibility that the target batch also has problems given the current evidence of deviation.

[0030] From this causal risk assessment model, causal chains with posterior probabilities greater than a preset threshold are extracted. These causal chains are then mapped back to the corresponding entities and relationship paths in the dynamic supply chain knowledge graph, generating information about the risk transmission path. To transform the probabilistic results into understandable action guidelines, causal chains with posterior probabilities greater than a preset threshold are extracted from the activated Bayesian network. The preset threshold is a configurable parameter used to filter out weak associations with extremely low probabilities. A causal chain is a sequence of multiple high-probability nodes and directed edges between them, such as air conditioning system failure → increased particle count in the filling area → decreased product aseptic assurance level. By obtaining the mapping relationship between Bayesian network nodes and entities in the dynamic supply chain knowledge graph, each significant causal chain is mapped back to the dynamic supply chain knowledge graph. The specific entities corresponding to each abstract node in the chain (such as a specific air conditioning unit, a specific filling line, or a specific product batch) and their actual relationships in the graph (such as being located, being processed, or belonging to) are found. This generates risk transmission path information, which clearly indicates which specific equipment, environment, or materials the risk passes through from the source of deviation, and with what probability and intensity, to which specific affected batches, providing targets for precise intervention.

[0031] This specification's embodiments utilize a Bayesian network model trained on historical data to transform implicit expert experience into a computable probabilistic model. This allows risk transmission analysis to be based on historical statistical patterns, significantly reducing subjectivity. The output path transmission strength quantifies the intensity of risk propagation along different paths using probability values, enabling decision-makers to distinguish between primary and secondary issues and prioritize limited emergency resources to block high-probability transmission paths. Furthermore, by mapping probabilistic causal chains to deterministic knowledge graph paths, the generated risk transmission path information combines statistical insight with the specificity of a knowledge graph. It not only answers how the risk might propagate and its probability but also clearly identifies which specific entities are involved in the transmission. This allows for precise handling of subsequent response measures (such as isolating specific equipment or recalling specific batches), avoiding economic losses from large-scale production stoppages or recalls.

[0032] Next, the risk index for each affected batch is calculated, specifically through the following method: Based on the deviation type, the pre-defined deviation severity mapping table is consulted to determine the corresponding baseline severity coefficient. According to the deviation type specified in the quality deviation emergency request (e.g., excessive endotoxin, visible foreign matter, or positive sterility test), the pre-defined deviation severity mapping table is consulted. This mapping table is a configuration table jointly developed by the quality assurance department and risk management experts based on the company's quality system, product risk level, and regulatory guidelines. It maps each possible deviation type to a corresponding baseline severity coefficient. This coefficient is a dimensionless numerical value reflecting the potential harm of that type of deviation to patient safety and product quality in an isolated situation. For example, a positive sterility test would be assigned a much higher baseline coefficient than a minor cosmetic defect.

[0033] From the risk transmission path information, all risk transmission paths pointing to the currently affected batch are extracted, and the path transmission strength corresponding to each risk transmission path is obtained. From the risk transmission path information generated in the previous steps, all risk transmission paths pointing to the currently affected batch are extracted. Each risk transmission path explicitly records a series of intermediate entities (such as shared equipment, adjacent storage locations) and relationships between the entity node of the deviated batch and the entity node of the currently affected batch. For each such path, its associated path transmission strength is read. This strength value is the posterior probability calculated through Bayesian network inference in the causal risk assessment step, quantifying the likelihood of successful risk transmission along that specific path. For example, a path passing through the same filling needle may have a higher transmission strength, while a path passing only through different areas of the same warehouse may have a lower transmission strength.

[0034] Based on the path type, the weight coefficient corresponding to each risk transmission path is retrieved from a pre-defined path weight configuration table. Based on the path type represented by each path in the map (e.g., direct shared equipment type, indirect personnel operation type, environmental media diffusion type), a pre-defined path weight configuration table is also retrieved. This configuration table is also pre-defined based on historical data and expert experience, assigning different weight coefficients to different types of transmission paths to reflect their relative importance in the specific context of biomedicine. For example, direct shared equipment types are generally considered to have higher risks and are therefore assigned a larger weight coefficient, while environmental media diffusion types may have a lower weight if the high-efficiency air filtration system is intact.

[0035] Obtain the production end timestamp of the affected batch and the production end timestamp of the deviation batch, calculate the absolute time difference, and determine the exponential decay factor. Obtain the production end timestamps of the currently affected batch and the deviation batch from the dynamic supply chain knowledge graph. These two timestamps are inherent attributes of the batch entity nodes, derived from the production completion records of the manufacturing execution system. Calculate the absolute time difference between them. Based on this time difference, apply a preset decay function (in this embodiment, it can be an exponential decay function) to calculate the exponential decay factor. The longer the production time interval between two batches, the lower the probability that they will be simultaneously affected by the same transient event (such as an occasional equipment failure). Therefore, the risk of the current batch should decay over time, simulating the dynamic decline of risk over time.

[0036] The risk index value for the affected batch is determined based on the severity baseline coefficient, the path transmission strength of each risk transmission path, the corresponding weight coefficient, and the exponential decay factor. The risk index value for the current affected batch is obtained by multiplying the sum of the products of the severity baseline coefficient, the path transmission strength of each path, and the corresponding weight coefficient, by the exponential decay factor calculated based on the time difference. This process integrates the severity of the deviation itself, the cumulative probability strength of risk transmission through multiple paths, the importance of different transmission methods, and the time buffering effect, thereby generating a comprehensive and dynamic quantitative risk score for each affected batch.

[0037] First, the inherent hazard of the deviation type itself is taken into account through a severity baseline coefficient, ensuring the objectivity of the assessment benchmark. By summarizing the transmission intensity of all exposure pathways and considering pathway weights, the dose of risk exposure is accurately measured. This allows two batches belonging to the same affected set to obtain drastically different risk values ​​due to differences in the number, type, and probability of exposure pathways, achieving differentiated and accurate profiling. Furthermore, the introduction of a time decay factor allows batches produced earlier or later to receive risk adjustments that reflect the actual situation, avoiding potential misjudgments that might arise from mechanical judgments based on fixed time windows.

[0038] Finally, the batch processing decision tree for the corresponding processing instructions is determined. The specific implementation process is as follows: The risk indicator value of each affected batch in the affected batch set is compared with a preset first risk threshold and a preset second risk threshold, where the first risk threshold is greater than the second risk threshold. First, the risk indicator value of each affected batch in the affected batch set is obtained; this risk indicator value is a normalized value calculated from the aforementioned risk quantification process. Then, this value is compared with two pre-set key parameters stored in the system configuration database: the preset first risk threshold and the preset second risk threshold, where the first risk threshold is numerically strictly greater than the second risk threshold. These two thresholds are not fixed but are determined by the company's Quality Risk Management Committee based on product characteristics, regulatory requirements, and historical quality data. They can be dynamically configured according to different product lines or deviation categories, essentially defining three tiered intervals for risk action.

[0039] Batches with risk index values ​​exceeding the first risk threshold are classified as scrap batches, and a first-class disposal instruction is generated. This first-class disposal instruction includes isolating inventory and initiating the scrapping process. This means that the risk level of the batch has exceeded the acceptable upper limit, indicating a high probability of quality defects with serious consequences. Therefore, a first-class disposal instruction is automatically generated. This instruction is a structured operation sequence that can be directly parsed and executed by downstream systems (such as warehouse management systems and quality management systems). Typical instructions include isolating inventory, where the system automatically marks the inventory status of the batch as quality isolated in the warehouse management system, triggers a physical transfer to the isolated area, and freezes all subsequent movement or shipping operations; and initiating the scrapping process, which automatically creates a deviation disposal order in the quality management system, associates it with the batch information, initializes an electronic scrapping and destruction process requiring approval, and simultaneously notifies relevant quality and warehouse personnel.

[0040] Batches with risk index values ​​between the second and first risk thresholds are classified as additional testing batches, and a second-type disposal instruction is generated. This second-type disposal instruction includes adding additional quality inspection items. These batches have clear risk exposure, but their severity and probability have not yet reached the critical point of immediate disposal. Therefore, a second-type disposal instruction is generated, the core of which is adding additional quality inspection items. The generation of this instruction is intelligent; it combines the type of deviation and the specific risk transmission path information that led to the batch being included in the affected batch set, and retrieves and matches targeted testing solutions from a pre-set enhanced inspection item mapping library. For example, for batches with deviations related to shared equipment paths and involving microbial contamination, the instruction may require increasing the aseptic sampling quantity for that batch or conducting specific microbial identification tests, rather than simply repeating routine inspections.

[0041] Batches with risk index values ​​below the second risk threshold are classified as conditionally released batches, and a third-category disposal instruction is generated. This third-category disposal instruction includes releasing the batch to the next process. This indicates that the batch risk is low, but monitoring is still necessary out of caution. The core operation of generating the third-category disposal instruction is releasing the batch to the next process. This instruction removes the pre-set pause status of the batch in the manufacturing execution system, allowing it to continue its flow. However, "conditional" means that a risk monitoring tag will be added to the batch's electronic production record, and a task will be automatically triggered, requiring operators in subsequent processes or online monitoring systems to conduct enhanced inspections or parameter monitoring.

[0042] The classification results and corresponding disposal instructions for the affected batch set are summarized and organized into a tree structure according to the batch hierarchy. The batch hierarchy is derived from a dynamic supply chain knowledge graph, such as the parent-child relationship between raw material batches, intermediate batches, and finished product batches. Using this relationship as the framework, each batch is treated as a node, with its risk classification, risk indicator value, and specific disposal instructions as node attributes, constructing a visualized batch disposal decision tree. The root node of the decision tree is usually the initial deviation batch, while the child nodes clearly show the scope of risk impact, the risk level of each node, and the system's recommended differentiated disposal paths, providing quality-responsible users with a global decision-making view and an automated operation entry point for one-click approval and execution.

[0043] First, by using pre-defined, scientifically verified risk thresholds, the previously vague qualitative judgments of high, medium, and low risk are transformed into automated classifications based on a unified standard, free from human ambiguity, ensuring the objectivity and fairness of decision-making standards. Second, specific and executable structured disposal instruction packages are pre-defined for each classification level, refining general requirements into specific actions such as isolating to a specific area and initiating an electronic disposal process numbered QA-101, or adding endotoxin testing and bacterial endotoxin checks. This significantly reduces the need for secondary interpretation and conversion after decision-making, avoiding execution deviations. Finally, by combining classification results with batch lineage relationships to generate a decision tree, a panoramic view of the risk impact and tiered response strategies is presented in an intuitive graphical format. This makes complex disposal plans clear at a glance, greatly improving the decision-making efficiency and confidence of quality managers. It also provides a clear and traceable electronic decision log for the entire disposal process, perfectly meeting the highest requirements of the biopharmaceutical industry for the rigor, consistency, and traceability of quality decisions.

[0044] Step S103: Based on the batch processing decision tree and the pre-stored supplier profile database, determine the corresponding supplier selection scheme and material allocation scheme to generate an emergency purchase order and realize emergency sourcing.

[0045] First, based on the batch disposal decision tree and the pre-stored supplier profile database, the corresponding supplier selection scheme and material allocation scheme are determined, specifically through the following methods: Based on the batch disposal decision tree, the material shortage and material categories caused by the scrapped batch and the additional inspection batch are calculated. First, the generated batch disposal decision tree is parsed, traversing all nodes marked as scrapped batches and additional inspection batches. For these nodes, the corresponding material category information (such as material code or common name) and planned consumption are extracted from the associated material entity attributes. By summing the consumption of the same material category for all related batches and comparing it with the current available inventory, the material shortage and specific material categories required to restore the production plan are calculated.

[0046] Using the material identifier corresponding to the material category as an index, the system retrieves a set of feasible suppliers from a pre-stored supplier profile database. Based on the time urgency requirement, this set of feasible suppliers is initially screened to determine the most viable suppliers. The system then queries the pre-stored supplier profile database using the identified material category as an index. This database is a dynamically maintained data warehouse, whose data is periodically updated via API from enterprise supplier master data, historical purchase orders, third-party business information platforms, and audit report systems, recording multi-dimensional information such as supplier qualifications, performance, and risks. The system initially retrieves a set of all feasible suppliers capable of providing the material by matching the material category with the supplier's supply capacity. To address the time pressure of urgent sourcing, the system performs an initial screening based on the implicit or explicit time urgency requirement of the emergency request (e.g., requiring materials to arrive within tens of hours). The screening logic includes checking whether the supplier's real-time operational status is normal and whether its historical average delivery time meets the urgency requirement, thereby excluding suppliers who clearly cannot meet the time requirements, forming a refined set of feasible suppliers to reduce the search space of subsequent optimization algorithms and improve computational efficiency.

[0047] Based on this, a multi-objective optimization model is constructed using the feasible supplier set as the decision variable, specifically including the following steps: For each feasible supplier in the feasible supplier set, a decision variable is defined to determine the sum of the products of the quantity of materials to be purchased from each feasible supplier and their corresponding unit price, in order to construct a goal of minimizing the total procurement cost. First, a mathematical representation is defined for the optimization problem, i.e., a decision variable is defined for each supplier in the feasible supplier set. This decision variable is a non-negative continuous or discrete variable, and its physical meaning represents the quantity of a specific material planned to be purchased from that supplier. For example, for supplier A, the decision variable X_A ​​represents the quantity of materials to be purchased from it.

[0048] Based on this, three objective functions need to be optimized simultaneously. The first objective is to minimize the total procurement cost. From the supplier profile database, the latest effective unit price for the currently required material category is extracted from each feasible supplier. This unit price can be a framework agreement price, a recent historical purchase price, or a supplier's real-time quote. Then, the total procurement cost is calculated by multiplying the decision variable (purchase quantity) for each supplier by its unit price, and then summing the products of all suppliers. This sum is the value of the total cost objective function that needs to be minimized.

[0049] The second objective is to minimize the latest delivery time target by obtaining the maximum promised delivery time from the set of feasible suppliers. This is achieved by retrieving the promised delivery time from the supplier profile database for each feasible supplier, based on current capacity, production schedule, and logistical feedback. The promised delivery time is expressed in either days or hours. For any procurement plan comprised of decision variables (i.e., selecting a set of suppliers and allocating procurement quantities), the latest delivery time is defined as the maximum promised delivery time among all selected suppliers. This maximum value represents the plan's performance in terms of delivery timeliness, and minimizing this value aims to ensure overall supply speed.

[0050] The third objective is to minimize supply chain risk, which relies on a dynamically calculated comprehensive risk score for each supplier. This comprehensive risk score is determined by integrating the quality certification status score of each viable supplier, the regional risk index corresponding to their geographical location, and the correlation between the viable supplier and the risk transmission path. The goal of minimizing supply chain risk is then constructed based on a weighted sum of the quantity of materials procured from each viable supplier and their corresponding comprehensive risk score. Specifically, the calculation of this score requires the integration of data from three dimensions. First, the quality certification status score, which is directly derived from the verified supplier qualification status (such as ISO 9001 certification, GMP audit pass) recorded in the supplier profile database, and is converted into a standardized score according to preset rules. Second, the regional risk index corresponding to the geographical location information, which is obtained by calling an external risk data service API, inputting the supplier's registered location or production plant address, and obtaining a comprehensive index of the region's real-time political stability, natural disaster risk, and epidemic control level. Third, the correlation between the feasible supplier and the risk transmission path information, which is a dynamically calculated indicator. The entity set describing the core technical characteristics, key raw material sources, or equipment list in the supplier profile is compared with the key entities involved in the risk transmission path information obtained in the previous analysis (such as specific faulty equipment models, specific pollution source areas). By calculating the set similarity or the association path analysis based on knowledge graph, the correlation score is obtained to quantify the supplier's proximity to this specific risk event.

[0051] This multi-objective optimization model is constructed by minimizing the total procurement cost, the latest delivery time, and the supply chain risk. The three sub-scores are then weighted and summed according to preset weights to obtain the comprehensive risk score for each supplier. Finally, the objective function for minimizing supply chain risk is defined as the sum of the products of all decision variables (procurement quantity) and their corresponding comprehensive risk scores for each supplier. This objective function guides the optimization model to tend to select the supplier combination with lower overall risk.

[0052] After defining the three objective functions mentioned above, a preliminary multi-objective optimization model is constructed. However, for the biopharmaceutical industry, optimization objectives alone are insufficient; rigid compliance constraints must be incorporated. Therefore, a constraint enhancement step is implemented. The first step involves extracting the real-time quality certification status of each feasible supplier from the supplier profile database. Based on this real-time quality certification status, a recertification time estimate is calculated for each feasible supplier. This status includes approved, pending audit, and unapproved states. The real-time quality certification status of each feasible supplier is extracted from the supplier profile database, typically categorized as approved, pending audit, and unapproved. Based on the company's internal "Supplier Management Procedures," a recertification time estimate is calculated for each state. For example, suppliers in the approved state do not require additional certification, and the estimated value is zero; suppliers in the pending audit state may need to complete supplementary documentation, and the estimated value is a shorter timeframe; new suppliers in the unapproved state require a complete on-site audit, sample testing, and other processes, and the estimated value is a longer timeframe. This estimated value represents the minimum time cost that must be reserved from the start of procurement to the compliant availability of materials.

[0053] The second step involves adding the recertification time estimate as a hard constraint to the multi-objective optimization model. This constraint ensures that the planned delivery time of materials purchased from the feasible supplier is later than the corresponding recertification time estimate. Specifically, for each supplier, the constraint states that the planned delivery time of materials purchased from that supplier (estimated based on promised delivery time and logistics time) must be later than (i.e., greater than or equal to) the supplier's recertification time estimate. This means that any procurement scheme that does not meet this constraint, such as attempting an urgent purchase from a new supplier but requiring a delivery time earlier than the time required to complete the audit, will be directly identified as an infeasible solution by the optimization algorithm, thus ensuring that all schemes output by the model are feasible in terms of regulatory procedures.

[0054] After constructing a multi-objective optimization model with the set of feasible suppliers as decision variables, and using the amount of material shortage to be filled as a constraint, the pre-constructed multi-objective optimization model is solved using a multi-objective optimization algorithm to obtain a Pareto optimal solution set. Each solution corresponds to a supplier selection scheme and a material allocation scheme. Based on the preset decision rules, the final solution is selected from the Pareto optimal solution set, and the supplier selection scheme and the material allocation scheme are determined according to the final solution.

[0055] Using the constraint of filling the material shortage, a multi-objective optimization algorithm (such as NSGA-II) is used to solve the constructed model containing multiple objectives and hard constraints. Due to the conflicting objectives, the algorithm generates a Pareto optimal solution set, where each solution represents a supplier selection and material allocation scheme that achieves different trade-offs among cost, time, and risk. Finally, a final solution is selected from this solution set based on preset decision rules. These rules can be user-defined preferences (e.g., prioritizing risk minimization) or selections based on more complex rules (e.g., selecting the solution with the highest weighted overall utility across the three objectives). Based on the specific values ​​of the decision variables in the selected final solution, the supplier selection scheme and the material allocation scheme are determined, generating a clear list of purchase orders. Finally, based on the supplier selection scheme and the material allocation scheme, an emergency purchase order is generated to achieve emergency sourcing.

[0056] This technical solution uses the recertification time estimate based on real-time quality status as a hard constraint embedded in the model, mathematically ensuring that all output solutions meet the most basic compliance requirements of the biopharmaceutical industry. It eliminates the short-sighted behavior of sacrificing regulatory compliance for speed, making the final sourcing solution no longer a compromise of a single objective, but a Pareto optimal solution after systematically balancing cost, timeliness, and risk under the premise of meeting rigid compliance. This ensures the scientific nature, compliance, and overall robustness of decision-making even in emergency situations, significantly improving the emergency response quality and long-term reliability of the supply chain.

[0057] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: By using a dynamic supply chain knowledge graph, data on materials, equipment, and the environment scattered across different systems are linked together. In the event of a deviation, all batches linked through production and logistics can be automatically and comprehensively traced, overcoming the ambiguity and lag of traditional manual investigation. A pre-trained causal risk assessment model is introduced to quantify risk transmission analysis. The model calculates risk transmission path information and risk indicators, transforming the binary judgment of whether an item is affected into a specific, quantifiable risk value for each batch. This allows subsequent handling to be based on objective data rather than subjective experience, achieving resource allocation based on precise risk measurement. Furthermore, a sourcing plan is generated based on a quantified batch handling decision tree and a supplier profile database, achieving intelligent connection from risk analysis to business action. This ensures that the starting point for sourcing decisions is the precise quantification of risk gaps, and inherently integrates the supplier's historical performance and compliance status (from the profile database). This allows emergency procurement not only to fill material gaps but also to proactively leverage the results of previous risk analysis, selecting supply sources that truly reduce system risk. Through these technical solutions, precise containment and scientific recovery can be implemented as quickly as possible after a quality deviation occurs, maximizing compliance with regulatory requirements and minimizing business interruption losses.

[0058] This specification also provides an emergency sourcing device for the biopharmaceutical supply chain, such as... Figure 2 As shown, the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method.

[0059] This specification also provides a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0060] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0061] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. An emergency source-finding method for a biomedical supply chain, characterized in that, The method includes: Receive a quality deviation emergency request containing a deviation batch identifier and deviation type. Based on the quality deviation emergency request, determine the deviation batch and the corresponding set of affected batches in a pre-built dynamic supply chain knowledge graph, and obtain the corresponding production-related data and logistics-related data. The production-related data includes time-series data of production environment sensors and production operation records, and the logistics-related data includes time-series data of transportation vehicle sensors and warehousing environment records. The production-related data and the logistics-related data are input into a pre-trained causal risk assessment model to determine the risk transmission path information from the deviation batch to the affected batches in the affected batch set, and the risk index of each affected batch is calculated to determine the batch disposal decision tree for the corresponding disposal instruction. Based on the batch processing decision tree and the pre-stored supplier profile database, the corresponding supplier selection scheme and material allocation scheme are determined to generate emergency purchase orders and realize emergency sourcing. 2.The biological medicine supply chain emergency source finding method according to claim 1, wherein, Based on the aforementioned quality deviation emergency request, before determining the batch with the quality deviation and the corresponding set of affected batches in a pre-constructed dynamic supply chain knowledge graph, the method further includes: Extract material batch entities, production process entities, equipment entities, and their supply relationship data, processing relationship data, and location relationship data from the enterprise resource planning system database, manufacturing execution system database, and warehouse management system as the first graph data source; From the IoT platform database, access the temperature sensor entity, differential pressure sensor entity, and particle counter entity associated with the device entity and the storage space entity, and obtain the environmental monitoring time series data generated by them in real time as the second map data source; Using an entity linking algorithm, the material batch entities and equipment entities in the first graph data source are aligned and merged with the corresponding sensor entities in the second graph data source to generate entity relationship pairs; The entity relationship pairs are associated with the Good Manufacturing Practice (GMP) clause entities obtained from the regulatory database, and compliance constraint attributes are added to the production process relationship edges in the graph. Entities and relationships with the aforementioned compliance constraint attributes are stored in a graph database to form the dynamic supply chain knowledge graph. 3.The biological medicine supply chain emergency source finding method according to claim 1, wherein, Based on the aforementioned quality deviation emergency request, the batches with quality deviations and the corresponding set of affected batches are identified in a pre-constructed supply chain knowledge graph, specifically including: Based on the deviation batch identifier, locate the corresponding deviation batch entity node in the dynamic supply chain knowledge graph; Based on the graph traversal strategy corresponding to the deviation type matching, the graph traversal strategy defines the time range threshold and the set of relationship types for querying entity relationships in the dynamic supply chain knowledge graph; According to the graph traversal strategy, starting from the deviation batch entity node, the traversal is performed along the supply relationship edge, the shared equipment processing relationship edge and the spatiotemporal proximity relationship edge to filter out all batch entity nodes within the time range threshold. All the selected batch entity nodes are combined with the in-transit batch entity nodes connected by logistics relationship edges to form the initial set of affected batches. Extract the key product quality attribute values ​​of each batch entity node in the initial affected batch set from the dynamic supply chain knowledge graph, compare them with the preset quality standard range, and remove the batch entity nodes whose key quality attribute values ​​meet the quality standard range to generate the final affected batch set. 4.The biological medicine supply chain emergency source finding method according to claim 1, wherein, The production-related data and the logistics-related data are input into a pre-trained causal risk assessment model to determine the risk transmission path information from the deviated batch to the affected batches in the affected batch set, specifically including: Extract historical deviation case data, including root cause annotations, from the historical deviation event database. The historical deviation case data includes historical production-related data, historical logistics-related data, and annotated root cause entities. The causal risk assessment model is determined by training a Bayesian network model using the historical deviation case data. In this model, network nodes represent different risk factor entities and deviation phenomenon entities, directed edges between network nodes represent causal relationships, and the weights of the edges represent conditional probabilities. The acquired production-related data and logistics-related data are transformed into input feature vectors defined by the causal risk assessment model, and input into the pre-trained causal risk assessment model to perform probabilistic inference. When the entity nodes of the deviation batch are used as evidence, the posterior probability of quality deviation of the entity nodes of the affected batches in the affected batch set is obtained, and the path transmission strength is determined. From the causal risk assessment model, extract the causal chains whose posterior probability is greater than a preset threshold, map the causal chains back to the corresponding entities and relationship paths in the dynamic supply chain knowledge graph, and generate the risk transmission path information.

5. The emergency source finding method of a biological medicine supply chain according to claim 1, wherein, Calculate the risk indicators for each affected batch, specifically including: Based on the deviation type, a preset deviation severity mapping table is queried to determine the corresponding severity baseline coefficient; From the risk transmission path information, extract all risk transmission paths pointing to the currently affected batch, and obtain the path transmission strength corresponding to each risk transmission path; Based on the path type, query the weight coefficient corresponding to each risk transmission path from the preset path weight configuration table; Obtain the production end timestamp of the affected batch and the production end timestamp of the deviation batch, calculate the absolute time difference, and determine the exponential decay factor. The risk index value of the affected batch is determined based on the severity baseline coefficient, the path transmission strength of each risk transmission path and the corresponding weight coefficient, and the exponential decay factor.

6. The emergency sourcing method for the biopharmaceutical supply chain according to claim 1, characterized in that, Determine the batch disposal decision tree for the corresponding disposal instruction, specifically including: The risk index value of each affected batch in the set of affected batches is compared with a preset first risk threshold and a second risk threshold, wherein the first risk threshold is greater than the second risk threshold; Batches with risk index values ​​greater than the first risk threshold are classified as scrapped batches, and a first type of disposal instruction is generated, which includes isolating the inventory and initiating the scrapping process. Batches whose risk index values ​​are between the second risk threshold and the first risk threshold are classified as additional testing batches, and a second type of disposal instruction is generated, which includes adding additional quality inspection items. Batches with risk index values ​​less than the second risk threshold are classified as conditionally released batches, and a third type of disposal instruction is generated, which includes releasing the batch to the next process. The classification results and corresponding disposal instructions of the affected batch set are summarized and organized into a tree structure according to the batch hierarchy to generate the batch disposal decision tree.

7. The emergency sourcing method for the biopharmaceutical supply chain according to claim 1, characterized in that, Based on the batch processing decision tree and the pre-stored supplier profile database, the corresponding supplier selection scheme and material allocation scheme are determined, specifically including: Based on the batch disposal decision tree, the material shortage and material categories caused by the scrapped disposal batch and the additional testing batch are calculated. Using the material identifier corresponding to the material category as an index, retrieve the set of feasible suppliers from the pre-stored supplier profile database, and perform preliminary screening of the set of feasible suppliers according to the time urgency requirements to determine the set of feasible suppliers; A multi-objective optimization model is constructed with the set of feasible suppliers as decision variables and the amount of material shortage to be filled as a constraint. The pre-constructed multi-objective optimization model is solved using a multi-objective optimization algorithm to obtain a Pareto optimal solution set, where each solution corresponds to a supplier selection scheme and a material allocation scheme. Based on preset decision rules, a final solution is selected from the Pareto optimal solution set, and the supplier selection scheme and the material allocation scheme are determined according to the final solution.

8. The emergency sourcing method for a biopharmaceutical supply chain according to claim 7, characterized in that, Constructing a multi-objective optimization model with the set of feasible suppliers as decision variables, specifically including: Define decision variables for each feasible supplier in the feasible supplier set, and determine the sum of the product of the quantity of materials purchased from feasible suppliers in the feasible supplier set and their corresponding unit prices, in order to construct a target that minimizes the total procurement cost; Obtain the maximum value of the committed delivery time from the set of feasible suppliers to construct a target that minimizes the latest delivery time; By integrating the quality certification status score of each feasible supplier, the regional risk index corresponding to the geographical location information, and the correlation between the feasible supplier and the risk transmission path information, a comprehensive risk score is determined, so as to construct a minimum supply chain risk target based on the weighted sum of the quantity of materials purchased from each feasible supplier and the corresponding comprehensive risk score. The multi-objective optimization model is constructed by minimizing the total procurement cost objective, the latest delivery time objective, and the supply chain risk objective. From the supplier profile database, extract the real-time quality certification status of each feasible supplier, and calculate the recertification time estimate for each feasible supplier based on the real-time quality certification status. The status includes approved status, pending audit status, and unapproved status. The estimated recertification time is added as a hard constraint to the multi-objective optimization model to constrain the planned delivery time of materials purchased from the feasible supplier to be later than the estimated recertification time corresponding to the feasible supplier.

9. An emergency sourcing device for the biopharmaceutical supply chain, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to perform the method as described in any one of claims 1-8.

Images